A government knowledge management system should run on the organization's own infrastructure, work without internet access where required, use local AI models, enforce record-level access, keep a tamper-evident audit log, handle scanned paper files, and support handovers when officers are transferred or retire. Ask vendors to demonstrate each of these with your own documents.
Why public bodies are different
- Data sensitivity: classified, personal and legally privileged records.
- Network rules: some networks have no internet access by design.
- Rotation: officers are transferred regularly, so knowledge leaves posts on a schedule.
- Paper: important records are often still on paper or scanned.
- Accountability: every access may need to be explainable later.
Requirements checklist
Deployment
- Runs on the organization's servers or a government cloud
- Can run fully air-gapped, with offline installation and updates
- AI models run locally; no calls to outside services
- Clear hardware requirements
Security and access
- Single sign-on with the organization's identity provider
- Record-level access rules (individual, team, department, organization)
- Tamper-evident audit log of access and changes
- Encryption at rest and in transit
- Clear data retention and deletion
Knowledge capture
- Handles scanned paper with local text recognition (OCR)
- Captures decisions and commitments, not only documents
- Works with meetings and dictation
- Respects personal privacy: work scope only
Transitions
- Shows what a departing officer holds
- Structured handover or exit interview
- Successor can ask questions with named sources
Procurement
- Source of AI models disclosed
- Data never used to train vendor models without consent
- Exit plan: export in open formats
Questions to ask vendors
- Show the system running with the network cable unplugged.
- Which AI model runs locally, and how good is it on our documents?
- How is the audit log protected from tampering?
- What leaves an employee's computer, and who can see it?
- How are updates delivered to an air-gapped network?
Where Reattend fits
Reattend can run on an organization's own server: seated computers then fetch models and updates from that server's mirror and talk to nothing else. Fully air-gapped rollouts, including the dashboard's AI, are scoped with each organization. Its desktop app captures work on each computer, including scanned paper read with the Mac's own text recognition, and only derived knowledge is shared under record-level access rules. The dashboard supports single sign-on via OIDC (Entra ID, Okta, Google Workspace; SAML on request) and keeps a hash-chained audit log. It shows topics only one person holds and runs exit interviews that become handoff documents. By default, Reattend's own model, Rabbit, runs on the computer; it is still in training, so evaluate it on your own documents. See security and our guide to air-gapped AI knowledge bases.
Frequently asked questions
Why do government bodies need on-premise knowledge management?
Many handle classified or sensitive records, have rules on where data may be stored, or operate networks without internet access. On-premise or air-gapped deployment keeps data inside the organization's infrastructure.
What should a government AI knowledge system include?
On-premise or air-gapped deployment, local AI models, record-level access control, single sign-on, a tamper-evident audit log, handling of scanned paper, and support for transfers and retirements.
Can AI work without internet access?
Yes, if the models run on the organization's own hardware. Capability depends on the model and hardware, so test with real documents before committing.